aurora-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kevin Sweeney (JIRA)" <>
Subject [jira] [Commented] (AURORA-1338) Remove ChainedPathDetector from thermos
Date Wed, 27 May 2015 21:34:48 GMT


Kevin Sweeney commented on AURORA-1338:

[~benley] I feel like that use case is better satisfied by the Vagrant environment - you can
already spin up a small mesos cluster to test with and launch tasks with the full Aurora config
syntax available, see how all the mesos isolator modules interact with your task, etc. Using
a standalone thermos you are restricted to a subset of those features.

There's also a lot of feature duplication between mesos - for example the CLI acts as an admin
control interface to the executor, but in mesos this is not really needed - you can use an
admin endpoint on the slave to kill tasks.

Similarly there's no reason to fsync checkpoints to disk if the observer is hosted in the
executor itself (and with the GC executor replaced by mesos task reconciliation).

PID namespaces obviate the logic that requires thermos to be really sure it's killing a process
it owns - as an executor in mesos it owns every process.

Disk isolation and resource accounting are also provided by Mesos now (though per-process
accounting is still useful).

Thermos had these features long before Mesos did, but IMO now that the core supports them
we should take advantage of them and reduce the amount of code we maintain.

> Remove ChainedPathDetector from thermos
> ---------------------------------------
>                 Key: AURORA-1338
>                 URL:
>             Project: Aurora
>          Issue Type: Story
>          Components: Executor
>            Reporter: Maxim Khutornenko
> Once the default checkpoint path moves into sandbox (AURORA-1337), remove the ChainedPathDetector
(1) and make sure FixedPathDetector (2) is not used in Aurora.
> (1) -
> (2) -

This message was sent by Atlassian JIRA

View raw message